In the design of wireless networks that feature Simultaneous Wireless Information and Power Transfer (SWIPT), a convenient mathematical simplification has long dominated the literature: the linear energy harvesting model. This model assumes that the direct current (DC) power harvested by a receiver is directly proportional to the radio frequency (RF) power it receives.
While mathematically elegant, this assumption violates basic circuit physics. Real-world energy harvesting circuits rely on physical diodes and rectifiers that exhibit highly non-linear behavior. Ignoring these non-linearities creates a significant gap between theoretical optimization and practical hardware performance. As a result, algorithms optimized under linear assumptions often fail in real-world deployments, overestimating harvested energy and leading to suboptimal resource allocation.
To bridge this gap, a growing body of research focuses on non-linear energy harvesting modeling. Understanding why these non-linearities occur and how they alter signal and network design is essential for building sustainable, self-powered wireless systems.
The Anatomy of RF-to-DC Non-Linearity
An RF energy harvesting circuit typically consists of an antenna, a matching network, and a rectifier (composed of diodes and capacitors) that converts the alternating RF signal into a stable DC voltage. The physics of these components dictate that the conversion efficiency is far from constant. Instead, the circuit operates across three distinct regimes based on the input power level:
Harvested DC Power
^
| /----------------- (Saturation Plateau)
| /
| /
| /
| / (Non-Linear Transition)
| /
| -------------------------/ (Sensitivity Threshold / Dead Zone)
+--------------------------------------------------------> Input RF Power
1. The Sensitivity Threshold (Dead Zones)
Physical diodes require a minimum activation voltage to turn on and begin conducting current. If the ambient RF signal is too weak, the induced voltage fails to cross this threshold, and the conversion efficiency plummets to zero.
Nikolaos I. Miridakis et al. characterize these regions as "dead zones" in Green Cognitive Relaying: Opportunistically Switching Between Data Transmission and Energy Harvesting. Similarly, in Wireless Networks with RF Energy Harvesting: A Contemporary Survey, Xiao Lu et al. note that conversion efficiency can drop from over 50% to less than 1% as the incoming signal weakens, meaning energy cannot be harvested at all if the ambient signal falls below the hardware's sensitivity limit.
2. The Saturation Plateau
Conversely, when the input RF power is exceptionally high, the physical components of the receiver circuit reach their current or voltage limits. At this point, any further increase in input power yields no additional harvested energy.
As Hei Victor Cheng et al. explain in Mixed-Timescale Beamforming and Power Splitting for Massive MIMO Aided SWIPT IoT Network, accounting for these circuit saturation points ensures that power-splitting strategies remain effective under high-intensity signal conditions where real-world hardware performance plateaus. Traditional linear models completely miss this effect, assuming that harvested power scales infinitely with input power.
Mathematical Formulations: Capturing the Curve
To capture these physical limitations without making optimization problems mathematically intractable, researchers use a variety of non-linear representational strategies.
Piece-Wise Linear and Sigmoidal Models
One common approach is to model the conversion efficiency using piece-wise linear functions or continuous sigmoidal (saturation-based) functions. In Enabling Full-Duplex and Energy Harvesting in Uplink and Downlink of Small-Cell Network Relying on Power Domain Based Multiple Access, Dinh-Thuan Do et al. demonstrate that these models prevent the overestimation of system performance at high signal-to-noise ratios (SNRs), providing realistic bounds for self-powered communication networks.
Higher-Order Polynomial Models
A more granular approach models the circuit's response by analyzing the Taylor expansion of the diode's current-voltage characteristic. This reveals that the harvested energy is not just a function of the average input power, but also depends on higher-order signal magnitudes.
In Simultaneous Information and Energy Transmission with Short Packets and Finite Constellations, Sadaf ul Zuhra et al. utilize a non-linear function that explicitly accounts for both the second and fourth powers of the signal magnitude. This higher-order modeling uncovers a fundamental intuition that fundamentally changes how we design wireless waveforms.
Why Waveform Design Flips: The Power of PAPR
The realization that rectifiers respond non-linearly to higher-order signal magnitudes completely upends traditional communication waveform design.
In standard data transmission, engineers prefer waveforms with a low Peak-to-Average Power Ratio (PAPR) to avoid distorting the signal in the transmitter's power amplifier. However, for energy harvesting, high PAPR waveforms are actually more efficient.
Low PAPR Waveform (Constant Power)
Voltage
^
| ---------------------------------- (Below Diode Turn-on Threshold)
|======================================
+----------------------------------------> Time
High PAPR Waveform (Bursty Peaks)
Voltage
^ __ __ __
| | | | | | | (Crosses Diode Threshold!)
| | | | | | |
|_____| |___________| |___________| |___
+----------------------------------------> Time
If a transmitter sends a smooth, constant-power signal (low PAPR), the voltage at the receiver might sit permanently below the diode's turn-on threshold, resulting in zero harvested energy. If the transmitter instead concentrates that same amount of total energy into short, high-intensity bursts (high PAPR), the voltage peaks will punch through the sensitivity threshold, activating the diode and allowing the receiver to harvest energy efficiently.
The authors of Simultaneous Information and Energy Transmission with Short Packets and Finite Constellations highlight this exact trade-off: waveforms optimized for energy transfer may intentionally deviate from those optimized for decoding reliability, as signals with high peak-to-average power ratios provide superior energy conversion efficiency despite being more difficult to decode.
Impact on Network Optimization
When moving from single-device physics to multi-user networks, non-linear modeling forces a complete redesign of resource allocation algorithms, beamforming strategies, and power-splitting policies.
Avoiding the Proximity Trap
In a multi-user SWIPT network, a base station beams power and data to multiple devices. Under a linear model, an optimization algorithm will naturally allocate a massive amount of power to devices closest to the base station, assuming their harvested energy will scale linearly to near-infinity.
However, Wenfeng Sun et al. in Joint spectral efficiency optimization of uplink and downlink for massive MIMO-enabled wireless energy harvesting systems demonstrate that this leads to severe resource allocation errors. Because real-world harvesting circuits saturate when devices are close to the power source, throwing extra power at near-field devices is entirely wasted. A non-linear framework recognizes this saturation plateau and intelligently redirects excess power to charge more distant devices or to boost data transmission rates.
Interference as an Asset
Non-linear modeling also changes how networks view co-channel interference. In traditional data networks, interference is a detriment. But in high-power environments, interference represents an additional source of harvestable electromagnetic energy.
As shown by Chao Shen et al. in Wireless Information and Energy Transfer in Multi-Antenna Interference Channel, incorporating non-linear saturation limits is vital when evaluating high-interference environments. Without these models, systems overestimate the energy captured from surrounding interference, leading to power-splitting and time-switching strategies that fail to maintain the energy loop required for self-powered IoT operations.
By replacing the linear illusion with high-fidelity non-linear models, the wireless community is moving toward optimization frameworks that accurately reflect physical hardware, paving the way for predictable, truly autonomous wireless-powered networks.